You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
NVIDIA Tegra Architecture
- Thread starter french toast
- Start date
-
- Tags
- nvidia
.The system benchmarks for the Xiaomi Mi3 are remarkably similar between the 1.8GHz Tegra 4 variant and the S800-8974AB 2.3GHz variant:
Antutu: 36582 vs. 36026
Quadrant: 18702 vs. 18975
Geekbench2: 4106 vs. 4139
That said, a GPU-centric benchmark should show some advantage for the latter SoC, but that data is not yet available on this smartphone platform.
Interestingly enough, these system benchmark scores for Tegra 4 in this smartphone are significantly better than what we saw from the Toshiba Excite Pro and HP Slatebook x2, and more or less match or exceed the numbers shown by NVIDIA on their reference Tegra 4 tablet.
As for why Xiaomi would use Tegra 4 in Mi3 for all their China Mobile customers, clearly Tegra 4 is a more cost-effective solution for this particular market than S800 (the latter which is more expensive due to an SoC die size that is ~ 50% larger in comparison), and the Mi3 will be selling for a surprisingly low price of $327 USD. FWIW, Spreadtrum is the company that is rumored to provide the baseband technology in the Mi3 Tegra 4 China Mobile version (the Icera i500 baseband modem will not finish qualification/certification testing until closer to end of this year).
Anyway, this is pretty big news for Tegra 4, as Xiaomi is one of the fastest growing companies in the world (and in history in fact), and China Mobile is the world's largest mobile service provider (not to mention the fact that Tegra 4 appears to have excellent benchmark performance in this smartphone form factor).
Antutu: 36582 vs. 36026
Quadrant: 18702 vs. 18975
Geekbench2: 4106 vs. 4139
That said, a GPU-centric benchmark should show some advantage for the latter SoC, but that data is not yet available on this smartphone platform.
Interestingly enough, these system benchmark scores for Tegra 4 in this smartphone are significantly better than what we saw from the Toshiba Excite Pro and HP Slatebook x2, and more or less match or exceed the numbers shown by NVIDIA on their reference Tegra 4 tablet.
As for why Xiaomi would use Tegra 4 in Mi3 for all their China Mobile customers, clearly Tegra 4 is a more cost-effective solution for this particular market than S800 (the latter which is more expensive due to an SoC die size that is ~ 50% larger in comparison), and the Mi3 will be selling for a surprisingly low price of $327 USD. FWIW, Spreadtrum is the company that is rumored to provide the baseband technology in the Mi3 Tegra 4 China Mobile version (the Icera i500 baseband modem will not finish qualification/certification testing until closer to end of this year).
Anyway, this is pretty big news for Tegra 4, as Xiaomi is one of the fastest growing companies in the world (and in history in fact), and China Mobile is the world's largest mobile service provider (not to mention the fact that Tegra 4 appears to have excellent benchmark performance in this smartphone form factor).
Those three benchmarks all getting not only almost exactly the same score but almost the same delta too (1.5% difference on first two, 0.8% on third) on very different SoCs is too coincidental to believe. Seems much more likely to me that someone misreported the SoC that was ran on one of them.
I doubt that, because the numbers came straight from Xiaomi's Mi3 launch presentation! See video here (starting at around 16 minutes): http://www.youtube.com/watch?v=Dc0HtojLL6I
You don't see how an official representative could screw this up? You don't remember when a Samsung rep said Galaxy S4 would use a Qualcomm S4Pro processor? You remember when they also said it'd be 1.2GHz A15 and 1.6GHz A7?
Your disregard here is a total slap in the face to anyone who takes technical analysis of processors seriously. It is really, really scientifically ignorant.
Your disregard here is a total slap in the face to anyone who takes technical analysis of processors seriously. It is really, really scientifically ignorant.
A slap in the face? Oh come on. Common sense dictates that Xiaomi's CEO and NVIDIA's CEO would have some understanding of benchmark details about their one new phone!?!
I would love for you to explain in technical terms why Tegra 4 and S800-8974AB cannot have very similar system [CPU+GPU+system] benchmark scores in these three benchmarks. If you can't do that, then you are just blowing hot air right now. Measured data is what it is. The proper way to analyze data is to measure it first, and then analyze it, not the other way around!
Anyway, we have no idea about the actual breakdown in scores for any of these benchmarks. Antutu and Quadrant measure CPU + GPU + Mem + system and provides a combined score. Geekbench measure a variety of different items and provides a combined score. Without knowing the detailed breakdown of each of these tests, and without knowing the actual multi-core CPU and GPU operating frequencies sustained during these tests, I don't see how we can do an extensive architectural analysis one way or another.
I would love for you to explain in technical terms why Tegra 4 and S800-8974AB cannot have very similar system [CPU+GPU+system] benchmark scores in these three benchmarks. If you can't do that, then you are just blowing hot air right now. Measured data is what it is. The proper way to analyze data is to measure it first, and then analyze it, not the other way around!
Anyway, we have no idea about the actual breakdown in scores for any of these benchmarks. Antutu and Quadrant measure CPU + GPU + Mem + system and provides a combined score. Geekbench measure a variety of different items and provides a combined score. Without knowing the detailed breakdown of each of these tests, and without knowing the actual multi-core CPU and GPU operating frequencies sustained during these tests, I don't see how we can do an extensive architectural analysis one way or another.
Last edited by a moderator:
A slap in the face? Oh please. Common sense dictates that Xiaomi's CEO and NVIDIA's CEO would have some understanding of benchmark details about their one new phone!?!
I would love for you to explain in technical terms why Tegra 4 and S800-8974AB cannot have very similar system [CPU+GPU+system] benchmark scores in these three benchmarks. If you can't do that, then you are just blowing hot air right now. Measured data is what it is. The proper way to analyze data is to measure it first, and then analyze it, not the other way around!
A healthy dose of skepticism is always appropriate with results that are not coming from an independent source. On this forum, people tend not to take manufacturers' claims at face value.
A healthy dose of skepticism is always appropriate with results that are not coming from an independent source. On this forum, people tend not to take manufacturers' claims at face value.
Agreed, but this data was clearly measured by the manufacturer. This does not appear to be a projection nor a guestimate nor an off-the-cuff remark by a random company representative. Anyway, I do believe that the data is what it is, but it should be obvious in one month's time when independent reviewers measure the device, so I'm going to leave it at that.
A manufacturer of a phone is hardly going to say, "We've got this amazing new phone which comes in two hardware configurations, depending on your network. One of them is much faster than the other, but hey, tough luck!"
I don't doubt that the two devices could be pretty close in performance but it seems a bit suspect that both should be exactly the same speed, regardless of the benchmark. Benchmark scores to tend to vary depending on the chipset, with certain chipsets doing better on one benchmark than another.
I don't doubt that the two devices could be pretty close in performance but it seems a bit suspect that both should be exactly the same speed, regardless of the benchmark. Benchmark scores to tend to vary depending on the chipset, with certain chipsets doing better on one benchmark than another.
As I also mentioned in the Qualcomm thread, the Tegra version of the Xiaomi MI 3 superphone scores very well at Gfxbench T-Rex, outperforming all of the Tegra 4 tablets and even the iPad. It's still a ways off the fan-cooled Shield, though.
After looking into a lot of videos of benchmarking and gaming done on S800 devices (particularly the Galaxy S4 LTE-A), I've been very disappointed to see that thermal throttling is not just relegated to extreme scenarios like multiple back-to-back benchmark endurance tests. It doesn't seem to show up within minutes of gaming like on that Tegra 4 Toshiba tablet, but it can apparently be triggered within a session of exploring the device's performance with a few heavy loads.
If this is the line SoC manufacturers want to try to toe with their high-end offerings, offering a level of performance that's unsustainable even within heavy-but-still-reasonable usage conditions, it's better that Apple doesn't even try to lead the top-end in performance considering the inconsistencies they'd be introducing to the gaming experience on iOS.
After looking into a lot of videos of benchmarking and gaming done on S800 devices (particularly the Galaxy S4 LTE-A), I've been very disappointed to see that thermal throttling is not just relegated to extreme scenarios like multiple back-to-back benchmark endurance tests. It doesn't seem to show up within minutes of gaming like on that Tegra 4 Toshiba tablet, but it can apparently be triggered within a session of exploring the device's performance with a few heavy loads.
If this is the line SoC manufacturers want to try to toe with their high-end offerings, offering a level of performance that's unsustainable even within heavy-but-still-reasonable usage conditions, it's better that Apple doesn't even try to lead the top-end in performance considering the inconsistencies they'd be introducing to the gaming experience on iOS.
@ Mariner, I don't think one can say that they are "exactly" the same speed, but the overall combined benchmark scores are remarkably similar. I'm sure if we were able to see a breakdown of the combined CPU+GPU+Mem+Sys benchmark scores, the individual items would not be exactly the same (even if close to each other in some instances). Certainly I will be curious to see the breakdown, and even more curious to see how much power is consumed by CPU vs. GPU for each Mi3 variant. Hopefully Anand will test that.
Agreed, but this data was clearly measured by the manufacturer. This does not appear to be a projection nor a guestimate nor an off-the-cuff remark by a random company representative. Anyway, I do believe that the data is what it is, but it should be obvious in one month's time when independent reviewers measure the device, so I'm going to leave it at that.
Every 6 months or so, GPU manufacturer A comes out with graphics card X and benchmarks showing that it performs 10% better than graphics card Y from manufacturer B. It usually turns out to be about the same in independent reviews.
Then 6 months later manufacturer B comes out with graphics card Z and similar claims, with similar results. That's just the way marketing works.
french toast
Veteran
The skepticism here is warrented, however im less interested in peak performance now as I am in sustained/average performance and software optimisation.
Ive noiced xiaomi has very good software engineers and tends to coax very good performance out of the available hardware, ive noticed this trend with other chinese manufacturers, it would be interesting to see tegra 4/s800 run on stock android and compare the performance to these mi3 phones, are they allowing higher frequencies on certain benchmarks ala samsumg? Or something else?
Ive noiced xiaomi has very good software engineers and tends to coax very good performance out of the available hardware, ive noticed this trend with other chinese manufacturers, it would be interesting to see tegra 4/s800 run on stock android and compare the performance to these mi3 phones, are they allowing higher frequencies on certain benchmarks ala samsumg? Or something else?
I think Xiaomi is pretty unique among IHVs for the way they deal with phones. Their continued investment in and development of MIUI with weekly updates is impressive.
It must be remembered that they started off without any phones at all but developed MIUI for a number of devices which enthusiasts then ported to many others (I used it on my old Motorola Defy and Atrix). A very well integrated operating system with some excellent bespoke apps. The move into hardware seems to have been successful for them but it's all based on their work on MIUI - it makes a nice change to have a hardware company which sees software support as something more than an inconvenience!
That said (and to bring my post back on topic!), it really does seem strange to release a high-end phone with what will be two different hardware layouts, especially when one of them (the Snapdragon 800) would appear to be capable of doing the job of supporting all of the networks around!
I wonder if there is some sort of fan situation for NV in China where a large number of users specifically want to have Tegra chips in their devices? Are NV still producing special higher-end versions of games with extra bells and whistles for their Tegra chips?
It must be remembered that they started off without any phones at all but developed MIUI for a number of devices which enthusiasts then ported to many others (I used it on my old Motorola Defy and Atrix). A very well integrated operating system with some excellent bespoke apps. The move into hardware seems to have been successful for them but it's all based on their work on MIUI - it makes a nice change to have a hardware company which sees software support as something more than an inconvenience!
That said (and to bring my post back on topic!), it really does seem strange to release a high-end phone with what will be two different hardware layouts, especially when one of them (the Snapdragon 800) would appear to be capable of doing the job of supporting all of the networks around!
I wonder if there is some sort of fan situation for NV in China where a large number of users specifically want to have Tegra chips in their devices? Are NV still producing special higher-end versions of games with extra bells and whistles for their Tegra chips?
D
Deleted member 13524
Guest
It must be remembered that they started off without any phones at all but developed MIUI for a number of devices which enthusiasts then ported to many others (I used it on my old Motorola Defy and Atrix). A very well integrated operating system with some excellent bespoke apps. The move into hardware seems to have been successful for them but it's all based on their work on MIUI - it makes a nice change to have a hardware company which sees software support as something more than an inconvenience!
AFAIK, the story goes differently.
MIUI started as a community/fan-made ROM, originally a team made of only chinese people.
What Xiaomi did was to employ that team to pay them to do what they were already doing voluntarily for fun, while keeping the ROM open-source (that might have been a condition made by the XIAOMI team). They just have to use the Xiaomi's phones as development devices, which isn't bad at all because the hardware they use is great.
The continuous software support is a welcoming heritage of the ROM team's origins in the fan-driven community, rather than the typical code-the-least-for-paycheck corporate attitude persists in some brands (most obvious being LG and pre-Google Motorola).
Even though I'm not fond of the ROM itself, I do envy the spectacular software support that a friend of mine has for his Mi2.
A slap in the face? Oh come on. Common sense dictates that Xiaomi's CEO and NVIDIA's CEO would have some understanding of benchmark details about their one new phone!?!
I would love for you to explain in technical terms why Tegra 4 and S800-8974AB cannot have very similar system [CPU+GPU+system] benchmark scores in these three benchmarks. If you can't do that, then you are just blowing hot air right now. Measured data is what it is. The proper way to analyze data is to measure it first, and then analyze it, not the other way around!
Anyway, we have no idea about the actual breakdown in scores for any of these benchmarks. Antutu and Quadrant measure CPU + GPU + Mem + system and provides a combined score. Geekbench measure a variety of different items and provides a combined score. Without knowing the detailed breakdown of each of these tests, and without knowing the actual multi-core CPU and GPU operating frequencies sustained during these tests, I don't see how we can do an extensive architectural analysis one way or another.
You need to start applying basic common sense filters to the information presented to you. Just stop and think about it.
There are a ton of variables in CPU and system performance. Getting them to all line up is very unlikely. I could explain to you lots and lots of different ways in which I expect these two CPUs to perform very differently depending on the test at hand, but if you understood these things you wouldn't even be making this argument. Now consider that you have benchmarks that test a bunch of crap including graphics (AnTuTu) and ones that test only CPU stuff (Geekbench) but they're still getting the same damn scores. Yes it's not impossible but the odds of this happening are extremely low, even via different subtests all happening to get appreciably different scores that manage to cancel each other out in that benchmark's weighting.
And here's the part where it's especially bad - we've seen lots of scores from Cortex-A15 and Krait 300 and we know Krait 400 is virtually the same uarch. And this peculiar phenomenon of scores lining up across multiple benchmarks hasn't happened before.
Seems pretty evident that you're perfectly willing to switch your brain off when looking at vendor claims. Here's a shocker - even companies screw up slides sometimes.
Ok, we will see next month who's intuition is right here, but I stand by my words because my common sense and sense of logic tells me that it would be completely unbelievable that two hands-on CEO's would spend several minutes talking about benchmark scores of their shiny new product and yet have zero clue about whether or not these benchmark scores match up to the devices in question.
Similar threads
- Replies
- 0
- Views
- 468
- Replies
- 9
- Views
- 1K
- Replies
- 90
- Views
- 16K
- Replies
- 21
- Views
- 9K
